CEP Group Meeting

Lorenzo Fabbri

ISGlobal

4/19/23

Overview

  • Aim: to research the short-term effects of postnatal exposures to non-persistent EDCs on neurodevelopment and neurobehavior in childhood, and how the metabolome and the proteome might mediate these effects.
  • How: by making use of the principles and criteria of triangulation (Lawlor, Tilling, and Davey Smith 2016). We will primarily rely on the use of MTPs (Muñoz and van der Laan 2012; Haneuse and Rotnitzky 2013; Díaz et al. 2021) in combination with TMLE (van der Laan, Benkeser, and Sofrygin 2018).
  • Population: the HELIX sub-cohort, consisting of \(\text{N}=1200\) mother-child pairs.
  • Exposures: non-persistent EDCs (phenols, phthalates, and organophosphate compounds), measured in childhood in a pool of two urine samples.
  • Outcomes
    • Raven’s Coloured Progressive Matrices, for assessing non-verbal intelligence.
    • Computerised n-back test, for assessing working memory.
    • Attention Network Test, to provide a measure of the efficiency of three different functions of attention.
    • Child Behavior Checklist, for assessing behavioural and emotional problems.
  • Mediators: serum and urine metabolome, plasma proteome.

The Research Questions

flowchart LR
  exposure(chemical) --> mediator(metabolite)
  mediator --> outcome(neurodevelopment)
  exposure --> outcome

Figure 1: Simplified DAG.

  • What are the short-term effects of childhood exposure to EDCs on the outcomes?

A ExWAS kind of analysis with the dependent variable \(y\) being the outcome, and the independent variable \(x_i\) being the levels of EDC \(i\).

  • What are the short-term effects of childhood exposure to EDCs on metabolites and proteins?

A XWAS analysis with the dependent variable \(m_i\) being the metabolite or protein \(i\), and the independent variable \(x_j\) being the levels of EDC \(j\).

  • What are the short-term effects of childhood exposure to metabolites and proteins on the outcomes?

A XWAS analysis with the dependent variable \(y\) being the outcome, and the independent variable \(m_i\) being the metabolite or protein \(i\).

Triangulation

flowchart LR
  paper[Paper 3] --> rq1("RQ1
                          neuro~chemical")
  paper          --> rq2("RQ2
                          met~chemical")
  paper          --> rq3("RQ3
                          neuro~met")
  rq1            --> rqReg("Observational study.
                            Method: Regression with MTPs.
                            Causal estimand: phi.
                            Uncertainty quantification: bootstrapping.")
  rq1            --> rq1outNeg("Outcome negative control.
                                Method: Regression with GLMs/GAMs.
                                Uncertainty quantification: bootstrapping.")
  rq1            --> rqLit("Literature search")
  
  rq2            --> rqReg
  rq2            --> rqCrossCohort("Cross-cohort comparison.
                                    Method: Regression with GLMs/GAMs.
                                    Uncertainty quantification: bootstrapping.")
  rq2            --> rqLit
  
  rq3            --> rqReg
  rq3            --> rqCrossCohort
  rq3            --> rq3mr("Mendelian Randomization.
                            Causal estimand: .")

Figure 2: Diagram summarizing research questions and methods.

  • Chemical → outcome
    • Multivariate analysis in observational data: application of multivariable regression to observational data (\(\phi^{\Delta, i} = \mathbb{E}[Y^{d_i}] - \mathbb{E}[Y]\)).
    • Outcome negative control study: aims to reproduce the same conditions as the real study, but using a different outcome not plausibly causally related to the exposure.
    • Literature search: aims to validate the obtained results with data from the published scientific literature.
  • Chemical → omic
    • Multivariate analysis in observational data
    • Cross-cohort comparison: compares results between two or more populations in different contexts that result in different confounding structures.
    • Literature search
  • Omic → outcome
    • Multivariate analysis in observational data
    • Cross-cohort comparison
    • Mendelian Randomization: instrumental variable is one or more genetic variant(s) that have been shown to robustly relate to the exposure.

Triangulation: Details

  • Regression analysis. We will use TMLE in combination with SL, and a library of estimators including both simple parametric models (i.e., glm), and data-adaptive semi-parametric models. Specifically, we will make use of the lmtp R package, in combination with TMLE and SL. We hypothesize the presence of residual confounding, especially due to genetic and parental factors, which would result in exaggeration of any true causal effect.
  • Outcome negative control study. We postulate that SE factors represent the main confounders for this association. We identified outcomes that are associated to SE factors, but that are not causally associated to the chemicals: having a car, and whether the child has their own room. We will thus test the assumptions regarding the relation between the SE factors with each outcome, and check whether these associations are in the anticipated direction.
  • Literature search. For each outcome ~ chemical association, we will perform a literature search to compare the obtained results.
  • Regression analysis. As above.
  • Cross-cohort comparison. We will compare the regression models’ results between subjects of different ethnic origins. We postulate that SE factors represent the main confounders for this association. Based on the associations between SE factors (e.g., maternal education) and the omic markers, we will compare the results of omic ~ chemical with the expected modified associations (e.g., weaker or stronger).
  • Literature search. As above.
  • Regression analysis. As above.
  • Cross-cohort comparison. We postulate that SE factors represent the main confounders for this association. We will thus test the association between SE factors and the omics, separately for the two sub-populations (e.g., white British and Pakistani in BiB (Wright et al. 2013)). If this association is smaller among Pakistani, if the association among white British were due to residual SE factor confounding, we would expect a weaker association among Pakistani.
  • MR. We will make use, ideally, of a weighted allele score of genetic variants known to be robustly associated with the omic markers as an IV. We will use methods, including sensitivity analysis, to explore the possibility of bias due to: (i) weak instruments, and (ii) violation of the exclusion restriction criteria. We will make use of 2SMR (Davey Smith and Ebrahim 2003; Pierce and Burgess 2013; Davey Smith and Hemani 2014). We will thus employ results from two different types of GWASs: a exposure GWAS, using The Metabolomics GWAS Server (Suhre et al. 2011; Shin et al. 2014), and a outcome GWAS, using the OpenGWAS project (Hemani et al. 2018; Elsworth et al. 2020).

Bibliography

Davey Smith, George, and Shah Ebrahim. 2003. “‘Mendelian Randomization’: Can Genetic Epidemiology Contribute to Understanding Environmental Determinants of Disease?*.” International Journal of Epidemiology 32 (1): 1–22. https://doi.org/10.1093/ije/dyg070.
Davey Smith, George, and Gibran Hemani. 2014. “Mendelian Randomization: Genetic Anchors for Causal Inference in Epidemiological Studies.” Human Molecular Genetics 23 (R1): R89–98. https://doi.org/10.1093/hmg/ddu328.
Díaz, Iván, Nicholas Williams, Katherine L. Hoffman, and Edward J. Schenck. 2021. “Nonparametric Causal Effects Based on Longitudinal Modified Treatment Policies.” Journal of the American Statistical Association 0 (0): 1–16. https://doi.org/10.1080/01621459.2021.1955691.
Elsworth, Ben, Matthew Lyon, Tessa Alexander, Yi Liu, Peter Matthews, Jon Hallett, Phil Bates, et al. 2020. “The MRC IEU OpenGWAS Data Infrastructure.” August 10, 2020. https://doi.org/10.1101/2020.08.10.244293.
Haneuse, S., and A. Rotnitzky. 2013. “Estimation of the Effect of Interventions That Modify the Received Treatment.” Statistics in Medicine 32 (30): 5260–77. https://doi.org/10.1002/sim.5907.
Hemani, Gibran, Jie Zheng, Benjamin Elsworth, Kaitlin H Wade, Valeriia Haberland, Denis Baird, Charles Laurin, et al. 2018. “The MR-Base Platform Supports Systematic Causal Inference Across the Human Phenome.” Edited by Ruth Loos. eLife 7 (May): e34408. https://doi.org/10.7554/eLife.34408.
Laan, Mark J. van der, David Benkeser, and Oleg Sofrygin. 2018. “Targeted Minimum Loss-Based Estimation.” In Wiley StatsRef: Statistics Reference Online, 1–8. John Wiley & Sons, Ltd. https://doi.org/10.1002/9781118445112.stat07908.
Lawlor, Debbie A, Kate Tilling, and George Davey Smith. 2016. “Triangulation in Aetiological Epidemiology.” International Journal of Epidemiology 45 (6): 1866–86. https://doi.org/10.1093/ije/dyw314.
Muñoz, Iván Díaz, and Mark van der Laan. 2012. “Population Intervention Causal Effects Based on Stochastic Interventions.” Biometrics 68 (2): 541–49. https://doi.org/10.1111/j.1541-0420.2011.01685.x.
Pierce, Brandon L., and Stephen Burgess. 2013. “Efficient Design for Mendelian Randomization Studies: Subsample and 2-Sample Instrumental Variable Estimators.” American Journal of Epidemiology 178 (7): 1177–84. https://doi.org/10.1093/aje/kwt084.
Shin, So-Youn, Eric B. Fauman, Ann-Kristin Petersen, Jan Krumsiek, Rita Santos, Jie Huang, Matthias Arnold, et al. 2014. “An Atlas of Genetic Influences on Human Blood Metabolites.” Nature Genetics 46 (6, 6): 543–50. https://doi.org/10.1038/ng.2982.
Suhre, Karsten, So-Youn Shin, Ann-Kristin Petersen, Robert P. Mohney, David Meredith, Brigitte Wägele, Elisabeth Altmaier, et al. 2011. “Human Metabolic Individuality in Biomedical and Pharmaceutical Research.” Nature 477 (7362, 7362): 54–60. https://doi.org/10.1038/nature10354.
Wright, John, Neil Small, Pauline Raynor, Derek Tuffnell, Raj Bhopal, Noel Cameron, Lesley Fairley, et al. 2013. “Cohort Profile: The Born in Bradford Multi-Ethnic Family Cohort Study.” International Journal of Epidemiology 42 (4): 978–91. https://doi.org/10.1093/ije/dys112.